Research On Robotic Binocular Vision Sensor Based On FPGA

With the continuous development of China's manufacturing industry,machine vision,especially the binocular vision that incorporates TOF technology,is developing at a high speed.Binocular vision sensors play an important role in various robotic arms,industrial control,real-time image processing and other fields.This topic combines the common camera with FPGA and TOF technology to design a low-cost binocular fusion TOF vision sensor based on FPGA.The system consists of the following modules.(1)Binocular camera module.The optical detection part of the module is based on the OV5640 chip,and the maximum pixel resolution is 2592*1944,which can be arbitrarily reduced.In this subject,it is 1280*720 pixels,and the total number of pixels is over 900,000.(2)FPGA logic module.This module takes EP4CE30F23C8 N chip as the core,designs camera driving logic,image acquisition and buffer logic,image preprocessing logic and image transmission logic,and transmits the preprocessed image data to PC through USB interface.(3)Storage module.This module uses a DDR2 chip,MT47H64M16 HR,and the capacity of each chip is 1GB.Two memory chips are used in this project,driven by the same control signal,and the main purpose is to expand the bandwidth.(4)USB transmission module.The chip used in this module is CYUSB3014,which is a highspeed integrated chip integrated with USB3.0 module,mainly used for image data transmission.(5)TOF module.This module uses the EPC-660 chip,which is an area array TOF depth image sensor with a resolution of 320*240 and the output frame rate is 30 frames.The function of the module is to combine with the results of binocular matching to eliminate mismatched points and improve accuracy.(6)PC host computer algorithm module.This module receives image data,realizes binocular image display and binocular matching algorithm,and combines TOF image data to improve the effect of binocular matching algorithm.This topic can realize the binocular image display of 17 frames per second with 1280*720 pixels,and realize the depth image display of 5 frames per second to calculate the distance of the nearest object in the image.The measurement range is 0.5 meter to 5.0 meter,within the effective range of 1.0 meter to 2.0 meter,the accuracy can reach less than 1cm.